Remodeling of the epithelial-connective tissue interface (ECTI) in oral epithelial dysplasia as visualized by noninvasive 3D imaging

Abstract

Early neoplastic features in oral epithelial dysplasia are first evident at the basal epithelium positioned at the epithelial-connective tissue interface (ECTI), separating the basal epithelium from the underlying lamina propria. The ECTI undergoes significant deformation in early neoplasia due to focal epithelial expansion and proteolytic remodeling of the lamina propria but few studies have examined these changes. In the present study, we quantitated alterations in ECTI topography in dysplasia using in vivo volumetric multiphoton autofluorescence microscopy and second harmonic generation microscopy. The label-free method allows direct noninvasive visualization of the ECTI surface without perturbing the epithelium. An image-based parameter, ‘ECTI contour’, is described that indicates deformation of the ECTI surface. ECTI contour was higher in dysplasia than control or inflammed specimens, indicating transition from flat to a deformed surface. Cellular parameters of nuclear area, nuclear density, coefficient of variation in nuclear area in the basal epithelium and collagen density in areas adjacent to ECTI were measured. ECTI contour differentiated dysplasia from control/benign mucosa with higher sensitivity and specificity than basal nuclear density or basal nuclear area, comparable to coefficient of variation in nuclear area and collagen density. The presented method offers a unique opportunity to study ECTI in intact mucosa with simultaneous assessment of cellular and extracellular matrix features, expanding opportunities for studies of early neoplastic events near this critical interface and potentially leading to development of new approaches for detecting neoplasia in vivo.

Introduction

Oral squamous cell carcinoma (OSCC) comprises 4% of cancers worldwide. OSCC has a mean five-year survival rate of approximately 60%, hardly unchanged over decades despite occurring in relatively accessible sites - the oral cavity and oropharynx (1). The five-year survival rate of patients with OSCC is highly correlated to stage of diagnosis, ranging from 20% in late stage to >80% if detected early. Oral epithelial dysplasia (OED) is often the precursor for OSCC (2) and it is now widely accepted that early detection and treatment of OED is key to improving patient prognosis (3, 4). Differentiation of OEDs from benign conditions, including inflammation, is needed to insure proper diagnosis and better clinical outcomes (5, 6). Current approaches which rely on visual inspection and biopsy fail to catch neoplasia at its earliest treatable phase, emphasizing the importance of discovering early indicators of neoplasia toward developing more effective detection methods.

Nonlinear optical microscopy methods of multiphoton autofluorescence microscopy (MPAM) and second harmonic generation microscopy (SHGM) have been used successfully to interrogate the subsurface in vivo cytologic and extracellular microenvironment of neoplastic epithelium (7–11), revealing cellular and extracellular alterations that are consistent with pathological examination. In pathological grading, OED is differentiated from normal oral mucosa by cellular and extracellular atypia as well as architectural alterations. A major architectural change observed in OED is focal expansion of epithelium due to uncontrolled epithelial growth and remodeling of lamina propria (LP) resulting in transformation of the epithelial connective tissue interface (ECTI), where the basement membrane (BM) is located. Second harmonic generation microscopy (SHGM) provides a unique contrast for fibrillar collagen in the LP and may be used with MPAM to model multilayered microstructure of tissues in a label-free manner. MPAM-SHGM from the same tissue volume enables rapid delineation of epithelium from LP since both layers have autofluorescence while the epithelium lacks SHG. The junction at which SHG signatures begin in depth due to the transition into LP represents the ECTI. While difficult to appreciate in transverse histological examination, the ECTI is a 3D surface. Volumetric imaging of epithelium using MPAM-SHGM with specificity to visualize the ECTI itself presents an opportunity to investigate how it is altered in OED.

ECTI plays an integral role in neoplasia, serving as the interface at which malignant cells cross the BM during invasion (12). The ECTI maintains a distinct separation between epithelium and underlying LP under normal conditions. In OED, the extracellular matrix (ECM) in LP and ECTI undergo significant remodeling that appears to support invasion. ECM remodeling in OED resulting from compression stress (13) and secretion of matrix proteases (14, 15) by neoplastic foci in the epithelium alters the equilibrium between synthesis and degradation of ECM components. These result in marked changes in BM/ECTI features including new ‘rete-like’ features (16) altering ECTI topography, a feature used in pathologic evaluation (16, 17). These topographical changes result from contributors including an expanding epithelium at focal sites of hyperproliferation (18), deregulation of ECM dynamics that alter matrix spatial organization (19), and possibly stiffness, inducing pockets of compliance (13).

ECTI has been studied under in vitro conditions using traditional methods of histology (20, 21) and electron microscopy (22, 23). Two studies involving animal models have examined the ECTI 2D shape in histological sections using fractal analysis showing increased irregularities in ECTI during OEDs vs. controls (12, 24). Study of ECTI as a 3D surface by electron microscopy allowed visualization of abnormalities in density of connective tissue papillae in humans and animal models of OSCC. No quantitative ECTI shape analysis was available. Electron microscopy requires removal of the epithelium and is unable to provide visualization beyond the ECTI into the LP.

For noninvasive delineation and study of the ECTI in oral mucosa in vivo, optical imaging presents interesting possibilities. Automated approaches have been applied to optical coherence tomography (OCT) 2D (x–z) B-scan images to separate epithelium from LP in studies by Lee et al. in human patients and by Pande et al. in a hamster model of neoplasia (25, 26). ECTI shape parameters were not evaluated, and though feasible, the 3D ECTI structure was not delineated. A challenge encountered with OCT and other reflectance based methods is that imaging relies on backscattered light for contrast, such that the nature of signal from tissue constituents in both the epidermis and LP is the same, resulting in relatively low contrast between epithelium and LP in oral mucosa. This challenge may be lessened in other tissues, such as skin in which melanin in the dermis allows greater contrast between epithelium and lamina propria (27). This was shown in a study in which the dermal-epidermal junction was delineated in 3D from confocal reflectance microscopy (CRM) images of human skin (28). In that study, delineation was improved in dark skinned patients having more melanin in the dermis than fair skinned patients. The study did not include evaluation of the dermal-epidermal junction in detecting neoplasia.

In a previous study, we reported on the use of in vivo MPAM-SHGM to delineate 2D cross-sectional morphology of the ECTI (akin to histology) in a hamster model for OED (29). An ECTI line-shape parameter (ΔLinearity, a measure of ECTI curviness) was defined and shown to increase from normal to OED. A distinct advantage of the study was the use of SHGM to specifically define the upper boundary of the LP at the ECTI, since SHG is not evident from epithelium and confined to the LP. While useful due to the noninvasive nature of imaging without fixation or destruction of epithelium, such analysis did not depict the ECTI as a 3D surface, necessary to aptly visualize pockets of matrix compliance in OED. In this study we examine the ECTI in 3D using MPAM-SHGM and present a new way to model the 3D ECTI surface, while examining cellular and extracellular atypia near regions of remodeled ECTI in OED. The parameter, ECTI contour, a measure of deformation of this surface, is introduced to assess this interface in control, OED, and inflammation.

Materials and Methods

Animal Model

Animal studies were approved by the Institutional Animal Care and Use Committee at the University of Texas Medical Branch and conformed to the Guide for the Care and Use of Laboratory Animals. OED was induced by topical treatment of 0.5% 9,10-dimethyl-1,2-benzanthracene (DMBA) in mineral oil on the left cheek pouch of four-week old male Golden Syrian Hamsters (Harlan Laboratories) three times a week for 8–12 weeks (25, 30). Controls were treated with mineral oil following the same procedure. This model used in many preclinical studies has histological and molecular similarities to human OED and OSCC (30, 31). In an additional group, sodium lauryl sulfate (SLS) was topically applied daily for four consecutive days to induce inflammation using a solution of 1.4% SLS, 29% calcium pyrophosphate and 18% glycerol in sterilized water (32, 33) and imaged on the 5^th day. Sixteen DMBA-treated, five SLS-treated and thirteen controls were used providing 24 OED, 17 inflamed and 41 control sites. Mild, moderate and severe dysplasia were combined into one OED group. Before imaging, animals were anesthetized with intraperitoneal injection of 100-mg/kg ketamine and 2.5-mg/kg xylazine. The buccal pouch was gently pulled out of the oral cavity and stretched onto a flat sample holder using pins then rinsed with PBS. Following imaging by MPAM-SHGM, imaged sites were biopsied and fixed in 10% neutral-buffered formalin then processed for histopathology by H&E staining.

Microscopy

The MPAM-SHGM system used for this study has been described previously (29, 34). Illumination was provided using a titanium sapphire (Ti:sapphire) femtosecond (~100fs, 82MHz) laser (Tsunami, Spectra Physics), tunable within 750–1000nm. For MPAM, 780nm excitation was used and auofluorescence emission was collected (450–650nm). This wavelength has been shown to excite endogenous fluorophores reduced nicotinamide adenine dinucleotide (NADH) and flavin adenine dinucleotide (FAD) (35). 840nm illumination was used for SHGM of fibrillar collagen in LP with signal collected through a 420/20nm bandpass filter. In all 82 sites, Z-stacks were obtained using 1μm z-steps and 0.625μm pixel size with a 40X 1.2NA C-Apochromat water immersion objective (field-of-view: 320×320μm) having a working distance of 185μm (z-interval of 2μm was used in some pilot acquisitions for visualization but not used in contour analysis). Z-stacks were composed of 8-bit grayscale images of 512×512 pixels and imaging was performed to approximately a depth of ~180μm. 28mW average excitation power was used with intensity control provided by a variable attenuator using a half wave plate and Glan-Thompson polarizer (Thorlabs, Newton).

Sites of interest for microscopy were identified on the buccal pouch following white light visualization on a widefield imaging system using an AF Micro NIKKOR 60mm 1:2.8 lens attached to a Nikon DS Fi1 camera. In DMBA animals, lesion sites characteristic of dysplasia ~1–2mm in size were identified and location noted for imaging. Similar locations were chosen from control and inflammation groups. Hamsters were positioned for MPAM-SHGM centering on identified sites of interest using a 10X, 0.3NA air objective (Plan-Neofluar) before the objective turret was switched to the 40x objective for image acquisition.

Data Analysis

MPAM-SHGM images were processed using ImageJ (NIH, Version 1.48u) and Imaris (Bitplane, Version 7.4.2) for 3D volume reconstructions. Because SHG is expected to occur only at this interface and beyond, the ECTI surface was defined as the boundary where SHG began in depth. For visualization, representative z-stacks from each group were median filtered (using a 2×2 matrix) and with adjustment in contrast/brightness in ImageJ to ease visualization of cytologic features, with the same contrast adjustment parameters applied to all z-stacks.

An ImageJ macro was developed to map and quantify ECTI contour from SHGM stacks. The ImageJ despeckle algorithm (median filter) was applied to each unprocessed image plane in the stack to reduce salt and pepper noise and Gaussian filtered to smooth the image (radius of 2 pixels). Each stack was considered an x,y,z matrix comprised of columns extending in z. Beginning at x,y=0,0 and progressing row by row, columns were sampled at each z for intensity to identify the depth where a positive SHG signal occurred (thereby defining the ECTI along each column). Positive SHG signal was determined based on a threshold intensity value. To insure this position represented the beginning of LP a ‘true-signal’ was considered only if three consecutive pixels along the z-direction were above-threshold intensity. The first was defined as the ECTI surface. After a complete scanning of all 262,144 columns for each stack the ECTI point nearest to the surface (the smallest depth value) was set as z=‘0’ in the local coordinate system creating a temporary reference plane from which each subsequent point on ECTI could be compared. An 8-bit grey scale depth-map was generated based on the distance between reference plane and each identified surface point on the ECTI (each pixel value represents a depth rather than intensity). Z-columns that did not contain signal in any z-depth were assigned a depth, z′=0 in the depth-map defining an interface at the surface. This created false surfaces for z-columns without any SHG signal, which is accounted for in a normalization step later in the process. Depth-maps containing signal in less than 40% pixels, considered to have poor SHG signal from LP, were excluded (four OED sites from the entire data set) leaving twenty four OEDs for further analysis. ImageJ plugin ‘3D Surface Plot’ was used to plot depth-maps representing the ECTI surface. ECTI surface area was calculated from depth-maps using Heron’s method for surface area calculation of a curved surface (36). The method divides a curvature joining four points into two identical triangles and calculates the surface area of each triangle. The sum of the two areas is estimated to be the surface area of the curvature joining the four points. Matrices of 2×2 pixels were defined for the entire depth-map and each matrix divided into two triangles sharing the same hypotenuse. Surface area of each of these triangles was calculated using Heron’s formula (36) with the cumulative curved surface area calculated by summing of all triangle areas. The calculated surface area was normalized by the area of the reference plane to generate an ‘ECTI contour’- a measure of the ECTI curvature.

Cytologic features including epithelial thickening, anisonucleosis, anisocytosis, overcrowding and/or discohesion of epithelial cells, nuclear and/or cellular pleomorphism and binucleation (37) were identified in MPAM images. Epithelial thickness was considered abnormal beyond 40μm, the typical thickness in this model (8, 29, 38). All other features were qualitatively assessed following guidelines presented in (37). The percentage of samples from control, OED and inflammation showing these features was calculated and presented in Table-1. Quantitative measurements were made for nuclear density, nuclear area and coefficient of variance of nuclear area (CoVa – a measure of anisonucleosis) in basal epithelium. Nuclear density was found by counting cells within regions of interest to determine number of cells per area, while nuclear area was found by manually delineating the borders of nuclei and determining area using ImageJ measurements tool. CoVa is represented as ratio of standard deviation over mean of nuclear size.

Table 1.

Visual assessment and distribution of cytological features associated with OED

Parameter	Defining Criteria	Normal (n=41)	Inflammation (n=17)	Dysplasia (n=24)
Epithelial Thickening	Shortest distance from keratinizing layer to ECTI	0	16	22
Anisonucleiosis/Anisocytosis	Abnormal variation in nuclear and cellular size (e.g. enlargement)	1	3	18
Overcrowding/Cell Discohesion	Nonuniform intercellular spacing	1	6	13
Pleomorphism	Atypical variation of nuclear and cellular shape	3	3	16
Binucleation	Two nuclei share single cytoplasm as a result of increased mitotic rate and defective cytokinesis	1	3	11

Collagen density in the LP was measured using a previously described method (39). SHGM z-stacks were thresholded in ImageJ using Otsu’s algorithm. The image plane with maximum positive signal (I) and two image planes 3μm above (I-3) and below (I+3) from each stack were selected for calculation. 246 images from control, inflammation and OED were analyzed. To restrict measurements to collagen and account for dark vessels in the field, binary masks were created for each image and subtracted from thresholded images. For comparison between groups, SHG density in the three planes across the full field of view was calculated for each sample (three image planes per sample) and an average determined per group. For evaluating spatial variation of collagen density, SHG signal density was calculated from regions of interests (ROIs) beneath compromised ECTI in the LP and neighboring LP in the OED group. 40 ROIs from compromised LP and 35 ROIs from neighboring LP were assessed from 24 OEDs.

Statistical Analysis

Statistical comparison between groups as determined by histopathology were performed using single factor ANOVA followed by Tukey’s post hoc test with p<0.05 considered significant (p-value <0.05 is represented by a single asterisk (*) and p<0.01 by a double asterisk (**)). Receiver operator characteristic curves (ROC) were generated using SAS software (SAS Institute Inc., Cary) to calculate area under the curve (AUC), sensitivity and specificity.

Results

3D volumes of MPAM-SHGM overlays for control and OED with corresponding histological sections are shown in Fig. 1. Normal buccal epithelium is characterized by a thin layer of keratinocytes (Fig. 1a, Left; z=4μm) at the surface followed by stratified squamous epithelium. Superficial epithelium (Fig. 1a, left; z=14μm), composed of mature cells with large round nuclei and low nuclear to cytoplasm ratio, and basal epithelium (Fig 1a, left; z=19μm) having smaller epithelial cells are shown in planes to the left. Cell nuclei in MPAM images appear as circular regions lacking fluorescence surrounded by autofluorescing cytoplasm. The final layer is LP (Fig 1a, left; z=46μm). The 3D multilayer volume reconstruction is shown in Fig. 1a, right. Transition between autofluorescence (Epithelium: magenta) and SHG (ECM: green) represents ECTI (Fig. 1a), the boundary between epithelium and LP.

Fig. 1. Layer resolved volumetric MPAM-SHGM of hamster oral mucosa showed cytologic and microstructural features associated with neoplasia.

Single optical sections and three-dimensional volumes of representative control (a), and dysplastic (b) hamster oral epithelium. x–y micrographs from MPAM (magenta) at different depth shows autofluorescence from keratinizing and epithelial layers. SHG (green) from collagen fibers in the lamina propria is shown in green. Histologic sections of control and dysplasia of corresponding ROIs are shown in respective insets. K: Keratinizing layer, SE: Superficial epithelium, BE: Basal epithelium, LP: Lamina propria. Scale bar, 50μm.

Fig. 1b shows OED with thickened keratinizing and epithelial layers and remodeled LP. Superficial and basal epithelial layers (Fig. 1b, left; z=100μm and 140μm respectively) showed a heterogeneous population of enlarged cells and nuclei and focal epithelial thickening resembling typical histological features of dysplasia (Fig. 1b; inset). Deeper planes in OED traverse across both basal epithelium and LP since focally thickened epithelium results downward bulging of ECTI. Such areas were not observed in control epithelium due to uniform epithelial thickness. Histologic sections of controls and OED showed similar features in cross sections. In histopathology of OED, focal epithelial thickening was seen along with nuclei of atypical size and shape (pleomorphism).

Figure 2 shows regions of interests (ROIs) from control and OED in basal epithelium adjacent to ECTI. Fig. 2a shows an LP SHGM image plane from control having dense thick fibers of collagen. Fig. 2b is a 3D surface mask of an SHG image stack from control tissue showing a flat topography at the top where basal epithelium meets LP. Fig. 2c and 2d shows a single image plane and a 3D surface mask respectively of a LP with OED. Areas devoid of SHG indicate collagen degradation/remodeling in the LP (Fig. 2c “*”) with pockets of matrix compliance in 3D surface masks (Fig. 2d “*”). Planes which traversed both LP and epithelium at the same depth were observed only in the OED group such as in Fig. 2e and 2f (top view), where basal epithelial cells with cytologic features consistent with OED are visible along with LP at the same plane. Collagen remodeling (Fig. 2e) is indicated by reduced density of collagen fibers and by the expanded epithelial boundary. Cytologic features of neoplasia such as cellular discohesion (Fig. 2f “white arrow”), anisonucleosis (Fig. 2e “*”) and atypical variation in cell shape (pleomorphism; Fig. 2f “black arrow”) are evident in same spatial location as the compromised LP.

Fig. 2. Label-free volumetric imaging revealed remodeling of ECM in OED and confirmed presence of neoplastic cells in areas of SHG voids.

a) SHGM of a single image plane from a control lamina propria; b) 3D surface map showing the flat topography of a control lamina propria comprising of 130 image planes from SHGM; c) SHGM of a single image plane from an OED lamina propria; d) 3D surface map showing downward bulging of the OED lamina propria comprising of 160 image planes from SHGM; “*” in (c) and (d) indicates areas of remodeled lamina propria; e) ROI of co-registered MPAM (magenta) and SHGM (green) from an OED showing presence of epithelial cells having neoplastic cytology (“*” and “→”) near remodeled lamina propria; f) en face view of the ROI shown in (e). Scale bar, 50μm.

Figure 3(a–c) shows ECTI surface maps from representative control, inflamed and OED tissues. The color scale (inset, Fig. 3a) represents distance of a particular pixel in the surface map from a reference plane at depth of z′=0. A surface with larger range of colors indicates more irregularities than one showing only few colors. ECTI surfaces in control and inflamed mucosa were relatively flat with average ECTI contour 1.52±0.35 and 2.22±0.88 respectively. ECTI from OED showed major alteration in its surface leading to an elevated ECTI contour (4.01±1.85). The increase in ECTI contour in OED was found to be statistically significant (**<0.01) from control and inflammation (Fig. 3d). Inflamed epithelium had a marginal increase from control in ECTI contour however was not statistically significant. Fig. 3e shows ROC curves comparing control and OED in the absence and presence of inflammation. The presence of inflammation reduced sensitivity from 91.7% to 87.5% and specificity from 90.2% to 87.9%. Overall, AUCs in absence and presence of inflammation were calculated to be 0.95 and 0.91 respectively.

Fig. 3. Remodeling of ECTI surface maps results in increased ECTI contour in OED.

ECTI surface maps generated from SHGM z-stacks for control (a), inflammation (b) and dysplasia (c) show changes in surface topography. Color scale in (a) indicate changes in depth of the ECTI surface map. Red indicates a point towards the top and blue indicates a point towards the bottom of the surface map. (d) ECTI contour calculated from ECTI surface maps for all three groups. (e) ROC curves of ECTI contour when inflammation is present (blue) or absent (green) in the control group with AUCs 0.91 and 0.95 respectively.

Visual assessment of several cytologic features from MPAM was performed to correlate presence or absence of cytologic abnormalities with changes in ECTI contour. As expected a large percentage of OEDs showed presence of at least three cytologic features listed in Table-1 and ECTI contour changes were accompanied by neoplastic features.

Representative examples of atypical cytology identified in MPAM images are shown in Figure 4(a–g). Cytologic abnormalities include presence of mitotic nuclei in superficial epithelium (Fig. 4a), nuclear-cellular pleomorphism and anisonucleosis (atypical shape and size of cells and nuclei; Fig. 4b–d, white arrows). An interesting feature observed was epithelial cells having single or multiple cytoplasmic finger-like processes (Fig. 4b–c “outline”). Clusters of neoplastic cells are also visible in Fig. 4d showing loss of normal cell polarity and variation in cell shape and size that are regarded as indicators of OEDs (40).

Fig. 4. Cytologic atypia found to be consistent with increased ECTI contour.

a–g) x–y micrographs from MPAM showing cytologic atypia in OED. Arrows and dotted outlines point towards specific features identified in in-vivo MPAM; h) association of cytologic features with ECTI deformation. Scale bar, 40μm.

Linear arrangements of epithelial cells forming Indian file-like patterns (Fig. 4f “arrow”) were identified similar to those reported in salivary gland neoplasms (41). Fig. 4g shows cells with varying degrees of cytoplasmic fluorescence intensities, potentially an indication of deregulated metabolic rates since the source of fluorescence is cytosolic NADH and FAD and a characteristic previously reported in OED (10). Fig. 4h shows observed relationship between deformations in ECTI contour and associated cytologic abnormalities. More than 90% of imaged sites showed abnormal cytologic features consistent with OED along with an increase in ECTI contour. Similarly, about 90% of the samples having normal cytologic features were found to have low ECTI contour.

Cytologic parameters of nuclear density, nuclear area and coefficient of variation (CoVa) in nuclear area near the ECTI were increased in OED when compared with control (Fig. 5a–c). Inflamed tissue was not significantly different from normal by any of three measures. Inflammation was statistically different from OED by nuclear area and CoVa, but not in nuclear density. ROC of basal nuclear density resulted in the lowest AUC (0.78), sensitivity and specificity, with inflammation reducing sensitivity by ~10% (Fig. 5d). Basal nuclear area showed reasonable accuracy (Fig. 5e) in differentiating OED from control with an AUC of 0.82 irrespective of presence or absence of inflammation. Of the three parameters, CoVa (Fig. 5f) had the highest accuracy (AUC=0.99) irrespective of the presence of inflammation as a potential confounding factor. Along with cytologic features, the OED group showed spatial variation of collagen (SHG) density in the LP across regions of compromised ECTI vs. surrounding areas (Fig. S1a). Additionally, assessment of total collagen density across three groups showed significantly lower values in OED vs. control and inflammation (Fig. S1b). Sensitivity and specificity of collagen density were found to be 94% and 92% respectively with an AUC of 0.98 (Table-2).

Fig. 5. Cytologic features of neoplasia measured from MPAM images.

Average basal nuclear density (a), basal nuclear area (b) and coefficient of variation (CoVa) of basal nuclear area (c) in control, inflammation and OED are shown. ROC curves for these parameters are shown in d, e and f respectively. Grey and black ROC curves are for data sets in presence and absence of inflammation as a confounding factor. “*” and “**” represent p< 0.05 and p<0.01 respectively.

Table 2.

Summary of the statistical analysis of ECTI contour and cellular features quantified from in-vivo MPAM-SHGM.

Groups		ECTI Contour	Basal Nuclear Density	Basal Nuclear Area	CoV of Basal Nuclear Area	Collagen Density
Control vs Dysplasia (n=65)	AUC (95% CI)	0.95 (0.87–1.00)	0.81 (0.71–0.92)	0.82 (0.72–0.93)	0.99 (0.97–1.00)	0.99 (0.98–1.00)
	Sensitivity (%)	91.70%	83.30%	79.2	95.80%	94.3%
	Specificity (%)	90.20%	70.70%	85.4	97.60%	94.4%
(Control+Inflammation) vs Dysplasia (n=82)	AUC (95% CI)	0.91 (0.83–0.99)	0.78 (0.68–0.88)	0.82 (0.72–0.93)	0.99(0.97–1.00)	0.98 (0.97–1.00)
	Sensitivity (%)	87.50%	70.80%	79.2	95.80%	94.3%
	Specificity (%)	87.90%	67.20%	81	98.30%	91.7%

Discussion

A noninvasive imaging method to evaluate three-dimensional ECTI topography in OED was presented in relation to neighboring cytologic and extracellular abnormalities. MPAM-SHGM allowed for the ECTI to be represented as a 3D surface expanding beyond limited views provided by traditional histopathology. ECTI in OED was characterized by downward bulging of the epithelium similar to a well-delineated pushing margin seen in pre-invasive OSCC (42) resulting in a contour change/deformation and quantified as ECTI contour. Results indicated delineation of OED from control/benign tissue in a hamster model of OED and inflammation. ECTI deformations were found to occur along with basal cellular atypia as measured by nuclear density, nuclear area and CoVa, and with collagen remodeling measured by collagen density.

Advantages of MPAM-SHGM over histology or electron microscopy include the ability for in vivo label-free imaging without tissue perturbation and the ability to evaluate cytologic, collagen, and ECTI features together in an intact mucosa. While histopathology provides assessment of cellular and architectural atypia together, it is typically in cross-section and requires tissue removal. With electron microscopy the 3D ECTI surface is appreciated however it requires removal of epithelium excluding assessment of cytologic abnormalities. Independent optical signatures arising from epithelium and fibrillar collagen in MPAM and SHGM, respectively, help in accurate segmentation of ECTI in 3D while preserving the mucosal structural integrity, important in assessing the interplay between epithelium and LP across the ECTI. This ability to extract independent signal contrasts is an advantage over reflectance based techniques, such as OCT and CRM discussed in the introduction, due to a clear contrast between epithelium (lacking SHG) and LP (strong SHG). While ECTI in oral mucosa may be segmented in OCT (26, 27), cytologic features are beyond the current capability due to resolution limits. Further study with CRM will be needed to examine ECTI demarcation in oral mucosa which is unpigmented and presents lower contrast between epithelium and LP than skin (particularly dark skin) in which dermal-epidermal junction has been delineated in 3D (43).

ECTI contour provided a measure of ECTI deformation found in OED. With increasing degree of remodeling and formation of new rete-like features in OED, the ECTI contour increased (Fig. 3). Inflamed tissue lacking OED did not display topographical alterations in ECTI contour since inflammation induced uniform epithelial expansion and little disruption of LP. However, average ECTI contour in inflamed tissue was higher than control (Fig. 3d) and may be attributed to vasculature near the surface of LP. Future studies may include vessel masks to blend vessel surfaces with ECTI surface and reduce artifacts in ECTI contour. Presence of inflammation in control group had a nominal effect on the sensitivity/specificity of ECTI contour (Table-2). Overall advantages of ECTI contour were that it allowed for assessment of deformation and the algorithm developed was fairly straightforward, using freely accessible software (ImageJ). The primary challenge in determining ECTI contour was identifying pixels with SHG signals indicative of collagen contribution with high accuracy which was ensured by requiring at least three consecutive depth locations showing a positive SHG signal. This algorithm is adaptable to other image stacks with a hard interface, as seen in SHGM. More advanced edge detection algorithms could be developed in future when detection of subtle changes in fluorescence intensities is necessary as has been done for OCT or CRM (25, 43). Given that depth of imaging encompasses the epithelium and LP, the algorithm could be readily applied in similar mucosal structures and in those in which rete-like ridges extend in depth. The general approach could also be applied in cases involving bulb-like rete ridges; in such cases each z-column may contain more than one transition, requiring an algorithm to detect each major transition as an edge.

Qualitative analysis revealed the vast majority of samples having ECTI deformation also showed cellular atypia. Three neoplastic cytologic features related to features in histology were identified from images in 96% OED samples. Importantly, in about 90% cases those features occurred along with increased ECTI contour (Fig. 4h), indicating presence of ECTI surface deformation whenever neoplastic cytology is present. Additionally, OED showed a significant decrease in collagen density coinciding with increased ECTI contour and consistent with expected collagen remodeling and possible altered matrix compliance. The link between these features and ECTI contour is anticipated due to known modulation of each other’s microenvironment. Since all stages of OED were grouped, it is possible that these percentages are on the lower end since changes in mild OED may be more difficult to identify in micrographs. Identification of cellular atypia and collagen remodeling accompanying ECTI deformation highlights the relevance of ECTI contour as a reliable indicator of OED. As expected, inflammation induced epithelial thickening but showed no significant change in ECTI contour, collagen density or evident cellular atypia in visual assessments.

ROC curves calculated with exclusion and inclusion of inflammation within the control group showed its effect on performance of each feature, tested since inflammation can pose complications in clinical diagnosis of OEDs and by optical methods (5) due to similarities in optical properties (6). Basal nuclear density and basal nuclear area showed reduced performance with inclusion of inflammation - markedly pronounced in the case of basal nuclear density (inflammation reduced sensitivity by ~10%). Presence of epithelial cells at different stages of maturity within the same layer in OED is reflected in increased CoVa in dysplasia. Inflamed epithelium on the other hand has a uniform growth rate, which is indicated in a low CoVa and lack of effect on performance of this parameter. Similarly delineation of OED from control and inflammation with collagen density as a parameter was shown. Performance of ECTI contour in delineation of OED from normal/inflamed mucosa was superior to nuclear density and area. ECTI contour showed sensitivity/specificity comparable to CoVa and collagen density with slightly lower AUC. However, 95% confidence interval of all three parameters overlapped significantly, indicating comparable differentiating power for OED despite inflammation.

As with the current clinical standard for diagnosis by pathology, it is likely that multiple parameters arising from the epithelium, ECTI and LP will need to be evaluated together to inform as to the pathological status of human oral mucosa. Statistical analysis of cellular atypia and collagen density in the basal epithelium showed high AUC for these measured features indicating a potential for multivariate classification with ECTI contour. CoVa and ECTI contour was correlated with CoVa and collagen density (Pearson correlation coefficient 0.65 and −0.71 respectively), which reduced reliability of multivariate linear regression algorithms. However, pairing of ECTI contour with basal nuclear density or basal nuclear area improved AUC, sensitivity/specificity over single parameters (Supplementary table-1). It will be of interest to explore multivariate combinations by principal component analysis or ten-fold cross validation possible with larger sample sizes. Cytologic, biochemical and microstructural abnormalities have been shown by MPAM-SHGM in neoplasia in animal models and humans (38, 44, 45). Careful investigation of these parameters collectively with cytologic and microstructural features presented here could provide a more comprehensive assessment of neoplastic alterations.

Investigating ECTI with cellular and extracellular atypia as potential diagnostic metrics would require additional effort, including demonstration in human mucosa. Sites in which the majority of OED and OSCC occur in humans are similar to mucosa studied here (e.g. ventral tongue and floor of the mouth). The objective lens chosen for this study was most appropriate for hamster buccal mucosa, however depths beyond the approximate 180 μm evaluated in this study would be needed for humans with epithelial thickening beyond this limit (e.g. high grade OED). SHGM has been performed on human oral mucosa (46) and feasible using longer working distance objectives. We anticipate that information gained from MPAM-SHGM could be used to develop improved diagnostic approaches either by other optical methods reliant on understanding of cellular/structural changes or even directly using MPAM-SHGM through endoscopy (47). The application of depth resolved optical imaging to human epithelial tissues (28, 48) including MPAM and SHGM (45, 46) and recent advancements that are pushing the depth limits of MPAM-SHGM and miniaturization (49) support the feasibility of MPAM-SHGM ECTI imaging in human tissue in the future.

It will be of interest to examine the application of ECTI contour to other sites of the oral cavity. Non-keratinized stratified squamous epithelium of the cheek and keratinized stratified squamous epithelium of the hard palate show rete-processes in the ECTI under normal condition (50). Although ECTI shape changes in these anatomical sites may not involve a dramatic transformation from flat to curved, progression of dysplasia often involves changes from regular to irregular rete shapes with longer and broader processes. We expect the novel surface mapping of ECTI introduced in this study will still be relevant since segmentation of ECTI will be possible in a non-invasive manner due to a transition in SHG at the interface.

To conclude, we show that ECTI topography changes from flat to curved surface, increasing surface area and thus ECTI contour in OEDs. We also showed cytologic features consistent with neoplasia in basal epithelium were associated with increased ECTI contour as was collagen remodeling. The significance of this work is that a new approach utilizing MPAM-SHGM for in vivo imaging has been shown for observing and quantifying ECTI topography in 3D without tissue perturbation. The ability to noninvasively study topographical changes at the ECTI along with cellular and extracellular atypia expands the possibilities for investigation of early neoplastic transformation in OED and could potentially lead to development of new diagnostic approaches.